Missing the documentation on how to use the GUI to train a model for in-silico prediction

[michaelsteidel86]

Dear AlphaPeptDeep developers,

I am wondering if I could train a model to predict CSS and RTs for diaPASEF based data analysis of samples including a non-common PTM.

Would have a FragPipe-derived ddaPASEF based library at hand, as well as some DIA-NN reports (based on legacy predictor or FragPipe library).

What are the essential steps required to train a predictor and use it for in-silico library prediction?
Which files are needed as input (not sure about what a "PSM" diann output (selectable in GUI) exactly refers to?
Are raw instrument data actually required to train the model (is a section of the "transfer" part, but *.d files are not a selectable option here.

Would highly appreciate any help / link to doc I am simply missing out here.

Thanks in advance!
Michael

[jalew188]

Hi @michaelsteidel86 AlphaPeptDeep does not support burker/sciex data for ms2 training as we don't have a good centroiding algorithms for TOF-MS. We plan to support this, hope it will not take too long

[jalew188]

You can try to use frag-pipe's output spectral library for MS2/RT/CCS training. In APD GUI, you can select speclib_tsv for both PSM and RAW

[michaelsteidel86]

Thanks. So I assume one really needs to specify RAW data here (which is actually a spectral library? Which of the options needs to be selected? ( as 'speclib_tsv' is not visible in most current APD GUI)